Customer Service

Complaint triage and first-draft response

Classify incoming complaints by type and urgency, then generate a first-draft response. Agent reviews, personalises, and sends. Faster turnaround, consistent tone.

AugmentationPattern-matchingTime savingCustomer experience

40–60%

reduction in average handling time

Opportunity assessment

Business Impact
2

Minor improvement. Small efficiency gain with limited effect on overall turnover or bottom line.

Feasibility
5

Very easy to implement. Ready-to-use tools exist. Can be up and running in days or weeks with minimal technical resource.

Data Readiness
4

Light data requirements. Uses straightforward inputs — documents, product descriptions, customer records — that are usually accessible with minimal prep.

Risk Exposure
4

Low risk. Limited external exposure. A human reviews output before it reaches anyone outside the team.

Change Complexity
4

Light people impact. A new tool slots into an existing workflow. Minimal training needed. Most people will adapt quickly.

Tooling required

Standard LLM

Things to consider

  • Klarna's AI assistant handled 2.3M conversations in its first month — two-thirds of all chats — doing the work of 700 full-time agents. Resolution time dropped from 11 minutes to under 2 minutes, with an estimated $40M profit improvement (Klarna).

  • Agent review before sending is non-negotiable — particularly for complaints with legal, safety, or reputational sensitivity.

  • Define the classification taxonomy first (complaint types, urgency levels, routing rules) — the AI applies your framework, it doesn't create one.

  • Train agents on how to work with AI drafts rather than starting from scratch: the habit shift is the main change management task.

  • Track two metrics from day one: agent acceptance rate (percentage of AI drafts sent with minimal edits) and customer satisfaction. If acceptance falls below 50%, the prompt or taxonomy needs work before scaling.

  • Experiment starter: Identify your three highest-volume complaint types. Use an LLM to classify and draft responses for 100 historical complaints of those types. Have agents rate the drafts blind. If 70%+ are rated usable with light editing, run a live pilot on those complaint types only.

AI Transformation Playbook

Ready to assess your own opportunities?

The playbook gives you the full 150-opportunity directory, scoring tools, and 100+ templates for every stage of an AI transformation programme.